Network
Design
So,
you've been tasked with designing the company's network (or
re-designing the old one). Thinkin' about puttin' in your two weeks
notice and taking a job that teaches you how to say catchy phrases
like "Do you want fries with that?" and "Thank you, please drive
through"? Don't sweat it, network design isn't that tough. But good
network design takes a little planning before you start running
cables. In this section I'll take you through the different topologies
(both physical and logical) and cover a few different design
philosophies. Here we go...
www.tartoos.com
Step One:
What is the network going to be used for now? Sharing simple
documents, or moving around large volumes of AutoCAD or graphics data?
Will you need desktop video conferencing? Or how about a file server
to store shared documents? Take a really good (realistic) look at what
you expect from the new network now.www.tartoos.com
Step Two:
Try to determine what the network will be used for one to two years
down the road. Is the company expanding? Are new "rat cages"
(cubicles) going up every week? You don't want to design a network
that's going to be obsolete soon after it's built...
Step Three:
Do you want every PC to have all of the company's applications loaded
on them (Thick Clients), or do you want everyone to load applications
from a server (Thin Clients). There are advantages and disadvantages
to both.
Thick Clients:
·
All applications are loaded on every PC.
·
Pros: If the network is down everyone can still use their PC's -they
just can't share their work.
·
Cons: If you have a large user base upgrading applications can be a
chore (especially if your user base is not very "computer literate").
Thin Clients:
·
All applications are loaded on an application server.
·
Pros: Very easy to upgrade applications -you only have to load new
software on the server.
·
Cons: If the network is down so is your company...
Personally I prefer Thick Clients -usually applications are not
upgraded often enough to warrant the possible downtime in a Thin
Client architecture. If you decide to go with a Thin Client network
consider RAID level five storage and redundant power supplies for your
servers -yes, you will want to have more than one server...
Step Three:
Draw out your network topology. There are two aspects to deal with:
Physical
Topology: Where
are the cables, switches, routers, etc going to go? Choose a central
location for all the servers to make managing them easier -this is
also where you want to start the physical layout of your network. Here
are a couple of physical topologies:
Buss
topology:
-
A linear network
with the server(s) at one end and the client PC's connected at
various points along the network -also referred to as a "backbone".
A buss topology offers the least amount of flexibility, but may be
just the ticket for a small LAN that isn't expected to expand much.
One of the big cons to a buss architecture is that if one section
goes down a large portion of your network could be affected.
Star
topology:
-
A network that
uses hubs or switches arranged in a spoke fashion around a central
hub or switch that all the servers are connected to. A star
architecture offers the most flexibility. If you need more
connections to the net just add another switch or hub close to the
location of the new PCs. Just be sure and have a spare switch or hub
on hand incase the connection point for the servers goes down...
www.tartoos.com
-
You could use a
combination of both: A buss topology that all the servers connect to
and a star topology going out to the client PCs. This type of
network design is popular if the network is very large or the
servers cannot be centrally located.
|
Logical
Topology:
Do you need more than one IP network? If you have fewer than 200
computers a single class "C" network will give you enough IP addresses
for the nodes you have now, plus some room to grow later. Don't
confuse the logical topology of your network with the different
departments in your company -each department does not need it's own
net. On the other hand if the folks in accounting do not need access
to the data and servers being used by the graphics arts department you
could put them on separate logical and physical nets -or use
VLANs.
My advice: Keep It Simple Stupid (KISS)! If you can design your
network within a single IP address space, then do it! Remember that
two different IP nets cannot communicate with each other directly -you
will need a router or a bridge if the different IP nets need to share
data. Be careful, you could design yourself into working 60 hour weeks
managing a network that is far more complex than it should be...
www.tartoos.com
Step Four:
Take one last look at the
network you have designed and what is expected of the network by the
people who are going to be using it. Make it clear that this is THE
LAST TIME that you will accept changes to the design of the new net.
Once you start laying cables and crimping connectors it is too late to
make any major changes to the network.
I would
put something in here about shared Ethernet networks vice switched
-but the cost of Ethernet switches has dropped so much that it doesn't
make sense to build a shared Ethernet network. I will tell you the
difference: www.tartoos.com
On
a shared Ethernet network all bandwidth is shared by every node
on the network. If one node is transmitting all others must wait for
the transmission to finish before they can compete for access to the
network. If two or more nodes transmit at the same time a collision
occurs and the data has to be retransmitted. For a very small net, or
if you are not moving large amounts of data around, a shared access
Ethernet network will perform just fine. A hub is used on
shared access twisted pair Ethernet networks because a hub just takes
data in from one node and sends it out to all of the devices that are
connected to the hub. Hubs are nothing more than signal amplifiers.www.tartoos.com
A
switch has a data buffer to temporarily store packets for delivery
between nodes. Each computer has its own dedicated bandwidth and never
has to wait for another node to finish transmitting data. Collisions
are never happen. Switches will also automatically learn the Media
Access Control (MAC) addresses that are connected
to them, so adding a new PC to the network is simple -just plug in the
cable and the switch will "see" the new connection and start sending
data to it. Ethernet switches are my first choice for any network
because the price is good and you can expand a switched Ethernet
network without adversely impacting performance. Some switches will
also allow you to run multiple lines between them to increase
bandwidth. If you're really smart you'll run the cables between
switches down different paths, that way if one cable breaks or is cut
by a clumsy building maintenance worker you still have some
connectivity between the switches. For more information on switches,
see the Switching 101 article. And that brings me to the last step:www.tartoos.com
Step Five:
Damage control. What do I need on hand to fix the net if it goes down?
Spare cable, connectors, Network Interface Cards (NICs), even a spare
switch are all good ideas. Try to get the bean counters to allocate a
separate budget for network hardware and scheduled upgrades (you don't
want a department head's new Pentium !!! PC to dig into the money for
your network do you? :) Don't run cable directly from a switch to the
desktop. Run the cable from the switch to a female connector enclosed
in it's own housing or a wall type outlet. Then run a cable from the
outlet to the PC. You can make a few spare short runs of cable
(preferably Cat five twisted pair). If there is a break in the
network, odds are it will be from the wall outlet to the PC -which you
can quickly replace with one of your spares.
Good luck!
|
مدرسة
الكمبيوتر 
